Abstract
Clustering in the key set is decreased by smoothing the key-to-address transformation, and by adding shadow buckets to an open chaining file. The keys are pre-hashed before the address division, to remove the effect of sequential properties in the key set. Shadow buckets in the key search sequence reduce the effect of nonuniformity in file loading, and decrease the number of maximum probes needed to locate a record. The combined effects of these techniques lead to improved file performance for secondary storage devices, as shown by empirical studies. © 1977, ACM. All rights reserved.